贝叶斯网络的结构学习综述

Overview of Bayesian Network Structure Learning

贝叶斯网络是一种描述变量间不确定性因果关系的概率图模型,广泛应用于预测、推理、诊断、决策风险及可靠性分析等领域。结构学习作为构建贝叶斯网络的基础,被证实为非确定多项式难题。文中将贝叶斯网络结构学习按照数据量大小分为完备数据和缺失数据,将完备数据下的贝叶斯网络结构学习分为近似学习算法和精确学习算法。根据上述分类方法,对现有算法及其相关的改进算法进行总结与分析对比。

Bayesian network is a probabilistic graphical model that describes the causal relationship of uncertainty among variables.It is widely used in prediction,reasoning,diagnosis,decision making risk,reliability analysis,etc.Structural learning,as the basis for building Bayesian networks,is proved to be a non deterministic polynomial problem.In this paper,Bayesian network structure learning is divided into complete data and missing data according to the amount of data.The Bayesian network structure learning under complete data is divided into approximate learning and precise learning.According to this classification,the existing algorithms and their improved algorithms are summarized,analyzed and compared.